7 challenges of large-scale autonomous driving safety deployment

The automotive industry is currently studying how to safely deploy autonomous driving on a large scale, and must be able to cope with various key challenges at the L4 and even L5 levels. In addition, the complexity of automotive electronic systems is also growing rapidly. While requiring computing power to keep pace with the growth in complexity, how to ensure the system's power consumption, heat dissipation, size, cost and safety are also important factors that automakers and solution providers need to consider.

7 key challenges

1. Cost

Some predict that if L4 and L5 autonomous vehicles are produced in 2020, the cost may be $75,000 to $100,000 more than a regular car. In fact, this number is too low considering the number of sensors required to achieve L4 and L5 autonomous driving. If the total cost exceeds $100,000, in order for these vehicles to be purchased, the price will need to drop significantly to make them affordable to consumers. Therefore, this will most likely mean that the first vehicles to be truly deployed for autonomous driving will be part of Mobility as a Service (MaaS), ride-sharing, or robot taxi fleets, establishing a new business model to support these more expensive vehicles by replacing the cost of human driving.

2 Can L3 really be deployed?

As shown in the figure above, L3 is the first step from ADAS to fully autonomous driving. It means that when conditions permit, the vehicle can complete all driving actions and has the function of reminding the driver. The driver does not need to monitor the driving environment and can be distracted, but cannot sleep. He needs to be able to take over the vehicle at any time to deal with possible situations that artificial intelligence cannot handle. This leads to an interesting phenomenon. Once the driver's hands are off the steering wheel, he will happily handle emails, texts, etc., so that his eyes and thoughts are also off the road. Once an accident occurs, how quickly can the distracted driver return to the "steering wheel"? From the perspective of the division of responsibilities, skipping L3 autonomous driving can better judge whether the driver is controlling the vehicle or the vehicle is driving autonomously. However, even if automakers decide to skip the L3 stage, the technical complexity required to go from L3 to L4 is much higher.

3 The drastic increase in computing requirements of sensors

Moving from ADAS to autonomous driving requires a more precise understanding of everything around the car, and to achieve this, the number of sensors on the car has increased dramatically, requiring multiple radars, on-board cameras, and lidar to essentially replace and enhance human eye perception. These sensors are not only expensive, but also require processor processing to understand what they "see" and changes in the situation outside the car, which is very different from the computing power required for simple ADAS functions such as adaptive cruise control or emergency braking.

4. Increased software complexity

With the development of technologies such as electrification, intelligence, and networking, the electronic and electrical architecture of autonomous vehicles has become more complex and will also face many problems, such as the increase in ECUs and big data, the higher complexity of the human-computer interface, the frequent hacking of connected cars, and the satisfaction of the differentiated needs of autonomous driving customers... It is understood that L5 autonomous driving vehicles are expected to require 1 billion lines of code, compared with the Boeing 787 Dreamliner, which requires only 14 million lines of code.

Therefore, the importance of software function virtualization and hardware simplification will be further enhanced, and this may become a reality in several forms. One is to integrate hardware into stacks for different latency and reliability requirements; the second is that a redundant "supercomputer" will replace the ECU; the third is to completely abandon the concept of control units and adopt intelligent node computing networks instead.

5 How to improve the acceptance of autonomous driving deployment

According to the latest data from the American Automobile Association (AAA), 73% of American drivers say they are afraid to ride in a fully autonomous car, and 63% of American adults believe it is unsafe to share the road with an autonomous car while walking or riding a bicycle. Only when drivers and passengers believe that safety is trustworthy enough will the public accept new advanced driver assistance systems (ADAS) and increasingly automated technologies.

Safety is a key guarantee for automotive electronic systems, and strict safety standards and certifications apply to anything that needs to maintain reliable performance when the driver demands it, such as braking, steering, etc. When we improve the autonomous decision-making of cars, we are essentially replacing human safety decisions with complex computer systems.

6 From prototype to mass production

当今自主决策计算系统原型通常是基于现有的服务器技术，其挑战在于尺寸、功耗和散热性不适合于汽车。所有这些特性都需要显著减少，人们普遍认为功耗需要减少10倍，尺寸需要减少5倍，如果两者都能实现，那么成本和散热将显著降低。这也将推动自动驾驶汽车在消费者领域和机器人出租车领域的真正部署。

7. Enhance the in-car passenger connectivity experience

For the terminal field, the V2X communication module that supports autonomous driving is used as the entry point to further integrate various functional modules in the car, including in-car computing, storage and other functions, to build a new in-vehicle electronics business field, which can also be said to be the business scope of chip companies. Creating higher-performance and safer connected cars, realizing autonomous driving faster, providing better in-car passenger interconnection experience, and becoming a software and hardware platform for managing in-car applications are the differentiated competitive advantages of car manufacturers.

How can Arm help overcome these challenges?

Automotive OEMs and Tier-1 manufacturers are increasingly aware of the need for a strong technology partner to help them solve these challenges, and they have unanimously identified Arm and its extensive automotive ecosystem as the right partner to achieve this goal. In fact, Arm has been working closely with the automotive industry to understand each of the above challenges and provide complete solutions to help large-scale deployment of autonomous driving.

1. Comprehensive coverage of computing needs

Arm CPUs and other IPs such as GPUs, ISPs, and NPUs allow Arm-based solutions to be used in various systems throughout the vehicle, and Arm's partners also offer the widest range of automotive-grade SoCs. This series of application processors (Cortex-A), real-time processors (Cortex-R), and small, low-power microprocessors (Cortex-M) are suitable for all stages of autonomous driving. As Arm's partners bring more computing power to heterogeneous SoC platforms, this will help meet greater computing power needs while reducing power consumption, price, size, and maintaining good heat dissipation characteristics.

2 Achieving “Safety Readiness” for Higher Levels of Automated Driving

The Safety Ready program includes Arm's existing safety products as well as new or future products, which are developed using rigorous functional safety processes, including system processes and development that support ISO 26262 and IEC 61508 standards. The Safety Ready program provides a one-stop software, tools, components, certifications, and standards to simplify and reduce the cost of integrating functional safety for Arm partners. With the Safety Ready program, partners and automotive OEMs can ensure that their SoCs and systems have the highest level of functional safety required for autonomous driving applications.

3. Performance and safety are both indispensable

Arm’s leadership in security does not stop at integrating the latest certifications and standards. Although the Split-Lock feature is not new to the industry, Arm is the first to introduce it to processors designed for high-performance automotive applications. Split-Lock technology is a disruptive security innovation that enables the following performance:

• Provides flexibility not previously possible with lock-step CPU deployments

• The CPU cluster in the SoC can be configured in “split-core mode” to achieve high performance, where two (or four) independent CPUs in the cluster can be used for various tasks and applications

• If configured in "Lockstep Mode", the CPUs will be in lockstep, creating a pair (or two pairs) of locked CPUs in the cluster for higher automotive safety integrity applications

• CPU clusters can be configured to run in either mode after silicon production

4 New “Automotive Enhanced” Products

In September 2018, Arm launched the first autonomous driving processor Cortex-A76AE with integrated functional safety, designed specifically for the automotive industry. The chip is equipped with Split-Lock technology, a disruptive safety innovation that has been implemented in an application processor for the first time.

However, for autonomous driving, the amount of real-time data collected and processed is much greater than that of ADAS. Therefore, in December 2018, Arm launched another IP for high data throughput computing, the Cortex-A65AE, which is aimed at the automotive electronics market and is mainly optimized for 7nm. Its biggest feature is that it supports SMT multithreading, and its performance throughput is 3.5 times higher than that of the previous generation. It is expected to be launched in 2020. At the same time, car companies may be willing to use the two chips, Cortex-A76AE and Cortex-A65AE, in combination to complete the perception process.

5. The broadest ecosystem advantage

As we all know, Arm is the IP supplier behind the chip. Currently, Arm has more than 1,500 IP licenses worldwide, and the shipment of chips based on the Arm architecture has reached 125 billion pieces. In 2017 alone, the shipment of chips based on the Arm architecture exceeded 21 billion pieces. Authorized partners include more than 500 industry leaders, startups, chip companies and OEM manufacturers. In addition, 16 top automotive chip manufacturers have authorized Arm's IP.